| Wireless sensor networks have been developing rapidly in recent years. It is across-discipline, being integrated embedded technology, sensors, distributed, wirelesscommunication and modern network; it is a new platform of obtaining information. Thecharacteristics of sensor networks determine that information fusion is hotspot of wirelesssensor networks research. Information fusion can achieve high-performance with low-cost inthe aspect of energy and communications, which is significant for sensor networks research.Firstly, in this paper, we start with resolving uncertain information of sensor network,and set up an information fusion model. After analysis of this model, we study intensively akind of directed probabilistic graphic model which is widely applied, and give itsrepresentation, principle and process of belief propagation. And then according to this, animproved algorithm is proposed for parameter learning, and a system of sensor network isbuilt up. Secondly, we test the performance of improved algorithm and do simulation to verifythe feasibility of the new model. The static probability graph model is proved validity fordealing with uncertain information.The static model cannot handle the objects which change state with time, so we need tofocus on dynamic models. In this paper, a distributed sequential Bayesian estimation model isproposed in which the belief state is transmitted in the wireless sensor networks, and isupdated using the measurements from the new sensor node. Due to the limitations of thepower and computational capability of sensor networks, belief representation is the key of themodel. For the dynamic model, two belief representation methods are proposed: Gaussiandensity approximation and a new LPG function approximation. By simulations,it has beentested on the performance of the dynamic models. Analyzed two methods, it has been shownthat the LPG function approximation algorithm has a wonderful performance in response timeand energy consumption. |
PDF Full Text Request